entrenar 0.7.10

Training & Optimization library with autograd, LoRA, quantization, and model merging
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46

//! Training hyperparameters configuration

use serde::{Deserialize, Serialize};

/// Training hyperparameters
#[derive(Debug, Clone, Serialize, Deserialize)]
#[serde(default)]
pub struct TrainingConfig {
    /// Number of epochs
    pub epochs: usize,
    /// Batch size
    pub batch_size: usize,
    /// Learning rate
    pub learning_rate: f64,
    /// Weight decay
    pub weight_decay: f64,
    /// Warmup ratio
    pub warmup_ratio: f32,
    /// Gradient accumulation steps
    pub gradient_accumulation_steps: usize,
    /// Maximum gradient norm
    pub max_grad_norm: f32,
    /// Enable gradient checkpointing
    pub gradient_checkpointing: bool,
    /// Mixed precision mode
    pub mixed_precision: Option<String>,
    /// Random seed
    pub seed: u64,
}

impl Default for TrainingConfig {
    fn default() -> Self {
        Self {
            epochs: 3,
            batch_size: 16,
            learning_rate: 2e-4,
            weight_decay: 0.01,
            warmup_ratio: 0.03,
            gradient_accumulation_steps: 1,
            max_grad_norm: 1.0,
            gradient_checkpointing: false,
            mixed_precision: None,
            seed: 42,
        }
    }
}